As a specialist in decision making in healthcare and life sciences, I developed this prediction model to improve the quality of decision making around clinical trials. They are expensive and trial failure can be costly. But the precursors of failure and success can largely be known and in this case the assumptions in trial design can be assessed and tested before resources are allocated.

While AI-driven trial simulation and synthetic patient modelling are now established, this outcome predictor is novel in formalising clinical trial prediction as a governance-grade decision system. Its novelty lies in making trial assumptions explicit, uncertainty traceable, and decisions replayable .

Importantly, these features are not what the existing commercial products offer. By analogy, they are offering and competing about building a better telescope. What this approach focus on is how people decide where to point the telescope. It is not a Silver Bullet predictor and will not tell you if the trial will succeed.

As a decision tool, this offers powerful new capabilities for trial decision makers. This is a decision infrastructure for clinical trials that makes risk, uncertainty, and assumptions explicit before capital is committed.

By converting trial designs into forward-looking probabilities of success and linking those probabilities to strategic outcomes it addresses a structural blind spot in drug development in which organisations invest hundreds of millions while being unable to state clearly what assumptions they are relying on, how uncertain those assumptions are, or what level of risk they are knowingly accepting.

An overview of the Predictor’s novel capabilities

While AI-driven clinical trial simulation, synthetic patient modelling, and probability-of-success estimation are now established in pharmaceutical R&D, this predictor is novel in its conceptual framing, system architecture, and decision logic.

The novelty lies in formalising clinical trial prediction as a decision system, offering the following:

1. Trial assumptions (baseline risk, effect size, uncertainty, endpoints, and success rules) are explicitly represented, versioned, and auditable, rather than being embedded implicitly within models or analyst workflows.
2. Unlike existing platforms where Probability of Success or PTRS is the final product, this system treats PoS as an intermediate construct that feeds structured decisions, including Assumption Cards and scenario-specific strategic summaries.
3. Scientific uncertainty modelling is strictly separated from downstream financial and portfolio interpretation (e.g., Expected Value), preventing commercial assumptions from contaminating probabilistic inference.
4. Each simulated decision is reproducible and replayable, enabling retrospective analysis of which assumptions were accepted, which uncertainties were tolerated, and how decisions would change under alternative assumptions.
5. The primary outputs are not statistical tables but governance information designed for cross-functional decision-making (clinical, regulatory, finance, portfolio), supporting organisational learning rather than single-use prediction.

Taken together, these features offer pharmaceutical companies, CROs and trial specialists a new decision infrastructure for clinical development, distinct from existing AI trial simulation and synthetic patient platforms.

The primary users who would value the results of this Predictor include

• Portfolio Committees and Investment Decision Bodies because ti forces explicit articulation of risk and uncertainty and reduces post-hoc rationalisation after failures. It offers decision traceability, explicit risk acceptance and portfolio-level consistency.
• Clinical Development Leadership at Phase II–III by clarifying what success actually means before execution of the trial, surfaces disagreements early to address baseline risk, effect plausibility, and supports structured go/no-go decisions. This is about scenario-based risk, success rules and assumption stress-testing.

What problem does the predictor solve?

It estimates the probability that a proposed clinical trial will meet its predefined success criteria, conditional on explicit assumptions about population, design, endpoints, and treatment effects.

Is this trying to predict whether a drug works, and isn’t the output assumption-dependent?

No. The predictor does not assert biological truth. It quantifies design-conditional uncertainty. Assumptions are explicit, inspectable, and stress-testable; dependence on them is a feature, not a flaw.

How is this different from standard biostatistics, Bayesian design, or Monte Carlo simulation?

The mathematics are familiar, but the novelty lies in systematising and exposing the entire reasoning chain that is usually implicit, fragmented, or siloed across teams.

What information does the predictor require?

Eligibility criteria, endpoint definitions, trial design parameters, baseline risk calibration, and a prior over treatment effects, including uncertainty.

How are synthetic patients used, validated, and justified?

Synthetic patients represent draws from a defined eligibility-constrained population distribution. They are not surrogates for individuals but instruments for propagating uncertainty through the design.

What does the output look like?

A probability distribution over trial outcomes and an estimated probability of success under the declared success rule, with accompanying diagnostics.

How should a low probability of success be interpreted?

As a signal of fragility or misalignment in the proposed design, not as an instruction to terminate development.

How does this differ from benchmarks and vendor PoS models?

Benchmarks extrapolate from historical averages. This predictor conditions on the specific proposed trial and exposes the assumptions driving its estimate.

For more information contact Dr Mike Tremblay, mike_tremblay@skythunder.net www.skythunder.net